Fedora kernel-2.6.17-1.2142_FC4 patched with stable patch-2.6.17.4-vs2.0.2-rc26.diff
[linux-2.6.git] / drivers / cpufreq / cpufreq.c
index 03b5fb2..29b2fa5 100644 (file)
@@ -4,6 +4,11 @@
  *  Copyright (C) 2001 Russell King
  *            (C) 2002 - 2003 Dominik Brodowski <linux@brodo.de>
  *
+ *  Oct 2005 - Ashok Raj <ashok.raj@intel.com>
+ *     Added handling for CPU hotplug
+ *  Feb 2006 - Jacob Shin <jacob.shin@amd.com>
+ *     Fix handling for CPU hotplug -- affected CPUs
+ *
  * This program is free software; you can redistribute it and/or modify
  * it under the terms of the GNU General Public License version 2 as
  * published by the Free Software Foundation.
@@ -23,6 +28,7 @@
 #include <linux/slab.h>
 #include <linux/cpu.h>
 #include <linux/completion.h>
+#include <linux/mutex.h>
 
 #define dprintk(msg...) cpufreq_debug_printk(CPUFREQ_DEBUG_CORE, "cpufreq-core", msg)
 
  * level driver of CPUFreq support, and its spinlock. This lock
  * also protects the cpufreq_cpu_data array.
  */
-static struct cpufreq_driver           *cpufreq_driver;
-static struct cpufreq_policy   *cpufreq_cpu_data[NR_CPUS];
+static struct cpufreq_driver *cpufreq_driver;
+static struct cpufreq_policy *cpufreq_cpu_data[NR_CPUS];
 static DEFINE_SPINLOCK(cpufreq_driver_lock);
 
-
-/* we keep a copy of all ->add'ed CPU's struct sys_device here;
- * as it is only accessed in ->add and ->remove, no lock or reference
- * count is necessary.
- */
-static struct sys_device       *cpu_sys_devices[NR_CPUS];
-
-
 /* internal prototypes */
 static int __cpufreq_governor(struct cpufreq_policy *policy, unsigned int event);
 static void handle_update(void *data);
-static inline void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci);
 
 /**
- * Two notifier lists: the "policy" list is involved in the 
- * validation process for a new CPU frequency policy; the 
+ * Two notifier lists: the "policy" list is involved in the
+ * validation process for a new CPU frequency policy; the
  * "transition" list for kernel code that needs to handle
  * changes to devices when the CPU clock speed changes.
  * The mutex locks both lists.
  */
-static struct notifier_block    *cpufreq_policy_notifier_list;
-static struct notifier_block    *cpufreq_transition_notifier_list;
-static DECLARE_RWSEM           (cpufreq_notifier_rwsem);
+static BLOCKING_NOTIFIER_HEAD(cpufreq_policy_notifier_list);
+static BLOCKING_NOTIFIER_HEAD(cpufreq_transition_notifier_list);
 
 
 static LIST_HEAD(cpufreq_governor_list);
-static DECLARE_MUTEX           (cpufreq_governor_sem);
+static DEFINE_MUTEX (cpufreq_governor_mutex);
 
-struct cpufreq_policy * cpufreq_cpu_get(unsigned int cpu)
+struct cpufreq_policy *cpufreq_cpu_get(unsigned int cpu)
 {
        struct cpufreq_policy *data;
        unsigned long flags;
@@ -90,20 +86,19 @@ struct cpufreq_policy * cpufreq_cpu_get(unsigned int cpu)
        if (!kobject_get(&data->kobj))
                goto err_out_put_module;
 
-
        spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
-
        return data;
 
- err_out_put_module:
+err_out_put_module:
        module_put(cpufreq_driver->owner);
- err_out_unlock:
+err_out_unlock:
        spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
- err_out:
+err_out:
        return NULL;
 }
 EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
 
+
 void cpufreq_cpu_put(struct cpufreq_policy *data)
 {
        kobject_put(&data->kobj);
@@ -132,7 +127,7 @@ static unsigned int debug_ratelimit = 1;
 static unsigned int disable_ratelimit = 1;
 static DEFINE_SPINLOCK(disable_ratelimit_lock);
 
-static inline void cpufreq_debug_enable_ratelimit(void)
+static void cpufreq_debug_enable_ratelimit(void)
 {
        unsigned long flags;
 
@@ -142,7 +137,7 @@ static inline void cpufreq_debug_enable_ratelimit(void)
        spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
 }
 
-static inline void cpufreq_debug_disable_ratelimit(void)
+static void cpufreq_debug_disable_ratelimit(void)
 {
        unsigned long flags;
 
@@ -157,7 +152,7 @@ void cpufreq_debug_printk(unsigned int type, const char *prefix, const char *fmt
        va_list args;
        unsigned int len;
        unsigned long flags;
-       
+
        WARN_ON(!prefix);
        if (type & debug) {
                spin_lock_irqsave(&disable_ratelimit_lock, flags);
@@ -204,14 +199,14 @@ static inline void cpufreq_debug_disable_ratelimit(void) { return; }
  *
  * This function alters the system "loops_per_jiffy" for the clock
  * speed change. Note that loops_per_jiffy cannot be updated on SMP
- * systems as each CPU might be scaled differently. So, use the arch 
+ * systems as each CPU might be scaled differently. So, use the arch
  * per-CPU loops_per_jiffy value wherever possible.
  */
 #ifndef CONFIG_SMP
 static unsigned long l_p_j_ref;
 static unsigned int  l_p_j_ref_freq;
 
-static inline void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
+static void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
 {
        if (ci->flags & CPUFREQ_CONST_LOOPS)
                return;
@@ -234,47 +229,53 @@ static inline void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci) {
 
 
 /**
- * cpufreq_notify_transition - call notifier chain and adjust_jiffies on frequency transition
+ * cpufreq_notify_transition - call notifier chain and adjust_jiffies
+ * on frequency transition.
  *
- * This function calls the transition notifiers and the "adjust_jiffies" function. It is called
- * twice on all CPU frequency changes that have external effects. 
+ * This function calls the transition notifiers and the "adjust_jiffies"
+ * function. It is called twice on all CPU frequency changes that have
+ * external effects.
  */
 void cpufreq_notify_transition(struct cpufreq_freqs *freqs, unsigned int state)
 {
+       struct cpufreq_policy *policy;
+
        BUG_ON(irqs_disabled());
 
        freqs->flags = cpufreq_driver->flags;
-       dprintk("notification %u of frequency transition to %u kHz\n", state, freqs->new);
+       dprintk("notification %u of frequency transition to %u kHz\n",
+               state, freqs->new);
 
-       down_read(&cpufreq_notifier_rwsem);
+       policy = cpufreq_cpu_data[freqs->cpu];
        switch (state) {
+
        case CPUFREQ_PRECHANGE:
-               /* detect if the driver reported a value as "old frequency" which
-                * is not equal to what the cpufreq core thinks is "old frequency".
+               /* detect if the driver reported a value as "old frequency"
+                * which is not equal to what the cpufreq core thinks is
+                * "old frequency".
                 */
                if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
-                       if ((likely(cpufreq_cpu_data[freqs->cpu])) &&
-                           (likely(cpufreq_cpu_data[freqs->cpu]->cpu == freqs->cpu)) &&
-                           (likely(cpufreq_cpu_data[freqs->cpu]->cur)) &&
-                           (unlikely(freqs->old != cpufreq_cpu_data[freqs->cpu]->cur)))
-                       {
-                               dprintk(KERN_WARNING "Warning: CPU frequency is %u, "
-                                      "cpufreq assumed %u kHz.\n", freqs->old, cpufreq_cpu_data[freqs->cpu]->cur);
-                               freqs->old = cpufreq_cpu_data[freqs->cpu]->cur;
+                       if ((policy) && (policy->cpu == freqs->cpu) &&
+                           (policy->cur) && (policy->cur != freqs->old)) {
+                               dprintk(KERN_WARNING "Warning: CPU frequency is"
+                                       " %u, cpufreq assumed %u kHz.\n",
+                                       freqs->old, policy->cur);
+                               freqs->old = policy->cur;
                        }
                }
-               notifier_call_chain(&cpufreq_transition_notifier_list, CPUFREQ_PRECHANGE, freqs);
+               blocking_notifier_call_chain(&cpufreq_transition_notifier_list,
+                               CPUFREQ_PRECHANGE, freqs);
                adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
                break;
+
        case CPUFREQ_POSTCHANGE:
                adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
-               notifier_call_chain(&cpufreq_transition_notifier_list, CPUFREQ_POSTCHANGE, freqs);
-               if ((likely(cpufreq_cpu_data[freqs->cpu])) && 
-                   (likely(cpufreq_cpu_data[freqs->cpu]->cpu == freqs->cpu)))
-                       cpufreq_cpu_data[freqs->cpu]->cur = freqs->new;
+               blocking_notifier_call_chain(&cpufreq_transition_notifier_list,
+                               CPUFREQ_POSTCHANGE, freqs);
+               if (likely(policy) && likely(policy->cpu == freqs->cpu))
+                       policy->cur = freqs->new;
                break;
        }
-       up_read(&cpufreq_notifier_rwsem);
 }
 EXPORT_SYMBOL_GPL(cpufreq_notify_transition);
 
@@ -303,22 +304,21 @@ static int cpufreq_parse_governor (char *str_governor, unsigned int *policy,
                return -EINVAL;
        } else {
                struct cpufreq_governor *t;
-               down(&cpufreq_governor_sem);
+               mutex_lock(&cpufreq_governor_mutex);
                if (!cpufreq_driver || !cpufreq_driver->target)
                        goto out;
                list_for_each_entry(t, &cpufreq_governor_list, governor_list) {
                        if (!strnicmp(str_governor,t->name,CPUFREQ_NAME_LEN)) {
                                *governor = t;
-                               up(&cpufreq_governor_sem);
+                               mutex_unlock(&cpufreq_governor_mutex);
                                return 0;
                        }
                }
-       out:
-               up(&cpufreq_governor_sem);
+out:
+               mutex_unlock(&cpufreq_governor_mutex);
        }
        return -EINVAL;
 }
-EXPORT_SYMBOL_GPL(cpufreq_parse_governor);
 
 
 /* drivers/base/cpu.c */
@@ -332,11 +332,11 @@ extern struct sysdev_class cpu_sysdev_class;
  * "unsigned int".
  */
 
-#define show_one(file_name, object)                                    \
-static ssize_t show_##file_name                                        \
-(struct cpufreq_policy * policy, char *buf)                            \
-{                                                                      \
-       return sprintf (buf, "%u\n", policy->object);                   \
+#define show_one(file_name, object)                    \
+static ssize_t show_##file_name                                \
+(struct cpufreq_policy * policy, char *buf)            \
+{                                                      \
+       return sprintf (buf, "%u\n", policy->object);   \
 }
 
 show_one(cpuinfo_min_freq, cpuinfo.min_freq);
@@ -345,6 +345,8 @@ show_one(scaling_min_freq, min);
 show_one(scaling_max_freq, max);
 show_one(scaling_cur_freq, cur);
 
+static int __cpufreq_set_policy(struct cpufreq_policy *data, struct cpufreq_policy *policy);
+
 /**
  * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
  */
@@ -363,7 +365,10 @@ static ssize_t store_##file_name                                   \
        if (ret != 1)                                                   \
                return -EINVAL;                                         \
                                                                        \
-       ret = cpufreq_set_policy(&new_policy);                          \
+       mutex_lock(&policy->lock);                                      \
+       ret = __cpufreq_set_policy(policy, &new_policy);                \
+       policy->user_policy.object = policy->object;                    \
+       mutex_unlock(&policy->lock);                                    \
                                                                        \
        return ret ? ret : count;                                       \
 }
@@ -401,8 +406,8 @@ static ssize_t show_scaling_governor (struct cpufreq_policy * policy, char *buf)
 /**
  * store_scaling_governor - store policy for the specified CPU
  */
-static ssize_t store_scaling_governor (struct cpufreq_policy * policy, 
-                                      const char *buf, size_t count) 
+static ssize_t store_scaling_governor (struct cpufreq_policy * policy,
+                                      const char *buf, size_t count)
 {
        unsigned int ret = -EINVAL;
        char    str_governor[16];
@@ -419,7 +424,14 @@ static ssize_t store_scaling_governor (struct cpufreq_policy * policy,
        if (cpufreq_parse_governor(str_governor, &new_policy.policy, &new_policy.governor))
                return -EINVAL;
 
-       ret = cpufreq_set_policy(&new_policy);
+       /* Do not use cpufreq_set_policy here or the user_policy.max
+          will be wrongly overridden */
+       mutex_lock(&policy->lock);
+       ret = __cpufreq_set_policy(policy, &new_policy);
+
+       policy->user_policy.policy = policy->policy;
+       policy->user_policy.governor = policy->governor;
+       mutex_unlock(&policy->lock);
 
        return ret ? ret : count;
 }
@@ -451,7 +463,7 @@ static ssize_t show_scaling_available_governors (struct cpufreq_policy * policy,
                        goto out;
                i += scnprintf(&buf[i], CPUFREQ_NAME_LEN, "%s ", t->name);
        }
- out:
+out:
        i += sprintf(&buf[i], "\n");
        return i;
 }
@@ -521,12 +533,12 @@ static ssize_t show(struct kobject * kobj, struct attribute * attr ,char * buf)
        policy = cpufreq_cpu_get(policy->cpu);
        if (!policy)
                return -EINVAL;
-       ret = fattr->show ? fattr->show(policy,buf) : 0;
+       ret = fattr->show ? fattr->show(policy,buf) : -EIO;
        cpufreq_cpu_put(policy);
        return ret;
 }
 
-static ssize_t store(struct kobject * kobj, struct attribute * attr, 
+static ssize_t store(struct kobject * kobj, struct attribute * attr,
                     const char * buf, size_t count)
 {
        struct cpufreq_policy * policy = to_policy(kobj);
@@ -535,7 +547,7 @@ static ssize_t store(struct kobject * kobj, struct attribute * attr,
        policy = cpufreq_cpu_get(policy->cpu);
        if (!policy)
                return -EINVAL;
-       ret = fattr->store ? fattr->store(policy,buf,count) : 0;
+       ret = fattr->store ? fattr->store(policy,buf,count) : -EIO;
        cpufreq_cpu_put(policy);
        return ret;
 }
@@ -562,7 +574,7 @@ static struct kobj_type ktype_cpufreq = {
 /**
  * cpufreq_add_dev - add a CPU device
  *
- * Adds the cpufreq interface for a CPU device. 
+ * Adds the cpufreq interface for a CPU device.
  */
 static int cpufreq_add_dev (struct sys_device * sys_dev)
 {
@@ -571,8 +583,15 @@ static int cpufreq_add_dev (struct sys_device * sys_dev)
        struct cpufreq_policy new_policy;
        struct cpufreq_policy *policy;
        struct freq_attr **drv_attr;
+       struct sys_device *cpu_sys_dev;
        unsigned long flags;
        unsigned int j;
+#ifdef CONFIG_SMP
+       struct cpufreq_policy *managed_policy;
+#endif
+
+       if (cpu_is_offline(cpu))
+               return 0;
 
        cpufreq_debug_disable_ratelimit();
        dprintk("adding CPU %u\n", cpu);
@@ -582,9 +601,7 @@ static int cpufreq_add_dev (struct sys_device * sys_dev)
         * CPU because it is in the same boat. */
        policy = cpufreq_cpu_get(cpu);
        if (unlikely(policy)) {
-               cpu_sys_devices[cpu] = sys_dev;
-               dprintk("CPU already managed, adding link\n");
-               sysfs_create_link(&sys_dev->kobj, &policy->kobj, "cpufreq");
+               cpufreq_cpu_put(policy);
                cpufreq_debug_enable_ratelimit();
                return 0;
        }
@@ -595,17 +612,17 @@ static int cpufreq_add_dev (struct sys_device * sys_dev)
                goto module_out;
        }
 
-       policy = kmalloc(sizeof(struct cpufreq_policy), GFP_KERNEL);
+       policy = kzalloc(sizeof(struct cpufreq_policy), GFP_KERNEL);
        if (!policy) {
                ret = -ENOMEM;
                goto nomem_out;
        }
-       memset(policy, 0, sizeof(struct cpufreq_policy));
 
        policy->cpu = cpu;
        policy->cpus = cpumask_of_cpu(cpu);
 
-       init_MUTEX_LOCKED(&policy->lock);
+       mutex_init(&policy->lock);
+       mutex_lock(&policy->lock);
        init_completion(&policy->kobj_unregister);
        INIT_WORK(&policy->update, handle_update, (void *)(long)cpu);
 
@@ -615,9 +632,36 @@ static int cpufreq_add_dev (struct sys_device * sys_dev)
        ret = cpufreq_driver->init(policy);
        if (ret) {
                dprintk("initialization failed\n");
+               mutex_unlock(&policy->lock);
                goto err_out;
        }
 
+#ifdef CONFIG_SMP
+       for_each_cpu_mask(j, policy->cpus) {
+               if (cpu == j)
+                       continue;
+
+               /* check for existing affected CPUs.  They may not be aware
+                * of it due to CPU Hotplug.
+                */
+               managed_policy = cpufreq_cpu_get(j);
+               if (unlikely(managed_policy)) {
+                       spin_lock_irqsave(&cpufreq_driver_lock, flags);
+                       managed_policy->cpus = policy->cpus;
+                       cpufreq_cpu_data[cpu] = managed_policy;
+                       spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
+
+                       dprintk("CPU already managed, adding link\n");
+                       sysfs_create_link(&sys_dev->kobj,
+                                         &managed_policy->kobj, "cpufreq");
+
+                       cpufreq_debug_enable_ratelimit();
+                       mutex_unlock(&policy->lock);
+                       ret = 0;
+                       goto err_out_driver_exit; /* call driver->exit() */
+               }
+       }
+#endif
        memcpy(&new_policy, policy, sizeof(struct cpufreq_policy));
 
        /* prepare interface data */
@@ -626,9 +670,10 @@ static int cpufreq_add_dev (struct sys_device * sys_dev)
        strlcpy(policy->kobj.name, "cpufreq", KOBJ_NAME_LEN);
 
        ret = kobject_register(&policy->kobj);
-       if (ret)
-               goto err_out;
-
+       if (ret) {
+               mutex_unlock(&policy->lock);
+               goto err_out_driver_exit;
+       }
        /* set up files for this cpu device */
        drv_attr = cpufreq_driver->attr;
        while ((drv_attr) && (*drv_attr)) {
@@ -644,12 +689,26 @@ static int cpufreq_add_dev (struct sys_device * sys_dev)
        for_each_cpu_mask(j, policy->cpus)
                cpufreq_cpu_data[j] = policy;
        spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
+
+       /* symlink affected CPUs */
+       for_each_cpu_mask(j, policy->cpus) {
+               if (j == cpu)
+                       continue;
+               if (!cpu_online(j))
+                       continue;
+
+               dprintk("CPU %u already managed, adding link\n", j);
+               cpufreq_cpu_get(cpu);
+               cpu_sys_dev = get_cpu_sysdev(j);
+               sysfs_create_link(&cpu_sys_dev->kobj, &policy->kobj,
+                                 "cpufreq");
+       }
+
        policy->governor = NULL; /* to assure that the starting sequence is
                                  * run in cpufreq_set_policy */
-       up(&policy->lock);
-       
+       mutex_unlock(&policy->lock);
+
        /* set default policy */
-       
        ret = cpufreq_set_policy(&new_policy);
        if (ret) {
                dprintk("setting policy failed\n");
@@ -657,10 +716,9 @@ static int cpufreq_add_dev (struct sys_device * sys_dev)
        }
 
        module_put(cpufreq_driver->owner);
-       cpu_sys_devices[cpu] = sys_dev;
        dprintk("initialization complete\n");
        cpufreq_debug_enable_ratelimit();
-       
+
        return 0;
 
 
@@ -673,12 +731,16 @@ err_out_unregister:
        kobject_unregister(&policy->kobj);
        wait_for_completion(&policy->kobj_unregister);
 
+err_out_driver_exit:
+       if (cpufreq_driver->exit)
+               cpufreq_driver->exit(policy);
+
 err_out:
        kfree(policy);
 
 nomem_out:
        module_put(cpufreq_driver->owner);
- module_out:
+module_out:
        cpufreq_debug_enable_ratelimit();
        return ret;
 }
@@ -695,6 +757,7 @@ static int cpufreq_remove_dev (struct sys_device * sys_dev)
        unsigned long flags;
        struct cpufreq_policy *data;
 #ifdef CONFIG_SMP
+       struct sys_device *cpu_sys_dev;
        unsigned int j;
 #endif
 
@@ -706,7 +769,6 @@ static int cpufreq_remove_dev (struct sys_device * sys_dev)
 
        if (!data) {
                spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
-               cpu_sys_devices[cpu] = NULL;
                cpufreq_debug_enable_ratelimit();
                return -EINVAL;
        }
@@ -715,25 +777,24 @@ static int cpufreq_remove_dev (struct sys_device * sys_dev)
 
 #ifdef CONFIG_SMP
        /* if this isn't the CPU which is the parent of the kobj, we
-        * only need to unlink, put and exit 
+        * only need to unlink, put and exit
         */
        if (unlikely(cpu != data->cpu)) {
                dprintk("removing link\n");
+               cpu_clear(cpu, data->cpus);
                spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
                sysfs_remove_link(&sys_dev->kobj, "cpufreq");
-               cpu_sys_devices[cpu] = NULL;
                cpufreq_cpu_put(data);
                cpufreq_debug_enable_ratelimit();
                return 0;
        }
 #endif
 
-       cpu_sys_devices[cpu] = NULL;
 
        if (!kobject_get(&data->kobj)) {
                spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
                cpufreq_debug_enable_ratelimit();
-               return -EFAULT;
+               return -EFAULT;
        }
 
 #ifdef CONFIG_SMP
@@ -757,7 +818,8 @@ static int cpufreq_remove_dev (struct sys_device * sys_dev)
                        if (j == cpu)
                                continue;
                        dprintk("removing link for cpu %u\n", j);
-                       sysfs_remove_link(&cpu_sys_devices[j]->kobj, "cpufreq");
+                       cpu_sys_dev = get_cpu_sysdev(j);
+                       sysfs_remove_link(&cpu_sys_dev->kobj, "cpufreq");
                        cpufreq_cpu_put(data);
                }
        }
@@ -765,18 +827,17 @@ static int cpufreq_remove_dev (struct sys_device * sys_dev)
        spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 #endif
 
-       down(&data->lock);
+       mutex_lock(&data->lock);
        if (cpufreq_driver->target)
                __cpufreq_governor(data, CPUFREQ_GOV_STOP);
-       cpufreq_driver->target = NULL;
-       up(&data->lock);
+       mutex_unlock(&data->lock);
 
        kobject_unregister(&data->kobj);
 
        kobject_put(&data->kobj);
 
        /* we need to make sure that the underlying kobj is actually
-        * not referenced anymore by anybody before we proceed with 
+        * not referenced anymore by anybody before we proceed with
         * unloading.
         */
        dprintk("waiting for dropping of refcount\n");
@@ -789,7 +850,6 @@ static int cpufreq_remove_dev (struct sys_device * sys_dev)
        kfree(data);
 
        cpufreq_debug_enable_ratelimit();
-
        return 0;
 }
 
@@ -825,7 +885,31 @@ static void cpufreq_out_of_sync(unsigned int cpu, unsigned int old_freq, unsigne
 }
 
 
-/** 
+/**
+ * cpufreq_quick_get - get the CPU frequency (in kHz) frpm policy->cur
+ * @cpu: CPU number
+ *
+ * This is the last known freq, without actually getting it from the driver.
+ * Return value will be same as what is shown in scaling_cur_freq in sysfs.
+ */
+unsigned int cpufreq_quick_get(unsigned int cpu)
+{
+       struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
+       unsigned int ret = 0;
+
+       if (policy) {
+               mutex_lock(&policy->lock);
+               ret = policy->cur;
+               mutex_unlock(&policy->lock);
+               cpufreq_cpu_put(policy);
+       }
+
+       return (ret);
+}
+EXPORT_SYMBOL(cpufreq_quick_get);
+
+
+/**
  * cpufreq_get - get the current CPU frequency (in kHz)
  * @cpu: CPU number
  *
@@ -842,12 +926,11 @@ unsigned int cpufreq_get(unsigned int cpu)
        if (!cpufreq_driver->get)
                goto out;
 
-       down(&policy->lock);
+       mutex_lock(&policy->lock);
 
        ret = cpufreq_driver->get(cpu);
 
-       if (ret && policy->cur && !(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) 
-       {
+       if (ret && policy->cur && !(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
                /* verify no discrepancy between actual and saved value exists */
                if (unlikely(ret != policy->cur)) {
                        cpufreq_out_of_sync(cpu, policy->cur, ret);
@@ -855,9 +938,9 @@ unsigned int cpufreq_get(unsigned int cpu)
                }
        }
 
-       up(&policy->lock);
+       mutex_unlock(&policy->lock);
 
- out:
+out:
        cpufreq_cpu_put(policy);
 
        return (ret);
@@ -869,7 +952,7 @@ EXPORT_SYMBOL(cpufreq_get);
  *     cpufreq_suspend - let the low level driver prepare for suspend
  */
 
-static int cpufreq_suspend(struct sys_device * sysdev, u32 state)
+static int cpufreq_suspend(struct sys_device * sysdev, pm_message_t pmsg)
 {
        int cpu = sysdev->id;
        unsigned int ret = 0;
@@ -897,7 +980,7 @@ static int cpufreq_suspend(struct sys_device * sysdev, u32 state)
        }
 
        if (cpufreq_driver->suspend) {
-               ret = cpufreq_driver->suspend(cpu_policy, state);
+               ret = cpufreq_driver->suspend(cpu_policy, pmsg);
                if (ret) {
                        printk(KERN_ERR "cpufreq: suspend failed in ->suspend "
                                        "step on CPU %u\n", cpu_policy->cpu);
@@ -931,14 +1014,14 @@ static int cpufreq_suspend(struct sys_device * sysdev, u32 state)
                freqs.old = cpu_policy->cur;
                freqs.new = cur_freq;
 
-               notifier_call_chain(&cpufreq_transition_notifier_list,
+               blocking_notifier_call_chain(&cpufreq_transition_notifier_list,
                                    CPUFREQ_SUSPENDCHANGE, &freqs);
                adjust_jiffies(CPUFREQ_SUSPENDCHANGE, &freqs);
 
                cpu_policy->cur = cur_freq;
        }
 
- out:
+out:
        cpufreq_cpu_put(cpu_policy);
        return 0;
 }
@@ -1012,7 +1095,8 @@ static int cpufreq_resume(struct sys_device * sysdev)
                        freqs.old = cpu_policy->cur;
                        freqs.new = cur_freq;
 
-                       notifier_call_chain(&cpufreq_transition_notifier_list,
+                       blocking_notifier_call_chain(
+                                       &cpufreq_transition_notifier_list,
                                        CPUFREQ_RESUMECHANGE, &freqs);
                        adjust_jiffies(CPUFREQ_RESUMECHANGE, &freqs);
 
@@ -1043,30 +1127,30 @@ static struct sysdev_driver cpufreq_sysdev_driver = {
  *     @nb: notifier function to register
  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
  *
- *     Add a driver to one of two lists: either a list of drivers that 
+ *     Add a driver to one of two lists: either a list of drivers that
  *      are notified about clock rate changes (once before and once after
  *      the transition), or a list of drivers that are notified about
  *      changes in cpufreq policy.
  *
  *     This function may sleep, and has the same return conditions as
- *     notifier_chain_register.
+ *     blocking_notifier_chain_register.
  */
 int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
 {
        int ret;
 
-       down_write(&cpufreq_notifier_rwsem);
        switch (list) {
        case CPUFREQ_TRANSITION_NOTIFIER:
-               ret = notifier_chain_register(&cpufreq_transition_notifier_list, nb);
+               ret = blocking_notifier_chain_register(
+                               &cpufreq_transition_notifier_list, nb);
                break;
        case CPUFREQ_POLICY_NOTIFIER:
-               ret = notifier_chain_register(&cpufreq_policy_notifier_list, nb);
+               ret = blocking_notifier_chain_register(
+                               &cpufreq_policy_notifier_list, nb);
                break;
        default:
                ret = -EINVAL;
        }
-       up_write(&cpufreq_notifier_rwsem);
 
        return ret;
 }
@@ -1081,24 +1165,24 @@ EXPORT_SYMBOL(cpufreq_register_notifier);
  *     Remove a driver from the CPU frequency notifier list.
  *
  *     This function may sleep, and has the same return conditions as
- *     notifier_chain_unregister.
+ *     blocking_notifier_chain_unregister.
  */
 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
 {
        int ret;
 
-       down_write(&cpufreq_notifier_rwsem);
        switch (list) {
        case CPUFREQ_TRANSITION_NOTIFIER:
-               ret = notifier_chain_unregister(&cpufreq_transition_notifier_list, nb);
+               ret = blocking_notifier_chain_unregister(
+                               &cpufreq_transition_notifier_list, nb);
                break;
        case CPUFREQ_POLICY_NOTIFIER:
-               ret = notifier_chain_unregister(&cpufreq_policy_notifier_list, nb);
+               ret = blocking_notifier_chain_unregister(
+                               &cpufreq_policy_notifier_list, nb);
                break;
        default:
                ret = -EINVAL;
        }
-       up_write(&cpufreq_notifier_rwsem);
 
        return ret;
 }
@@ -1115,35 +1199,36 @@ int __cpufreq_driver_target(struct cpufreq_policy *policy,
                            unsigned int relation)
 {
        int retval = -EINVAL;
+
        lock_cpu_hotplug();
        dprintk("target for CPU %u: %u kHz, relation %u\n", policy->cpu,
                target_freq, relation);
        if (cpu_online(policy->cpu) && cpufreq_driver->target)
                retval = cpufreq_driver->target(policy, target_freq, relation);
+
        unlock_cpu_hotplug();
+
        return retval;
 }
 EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
 
-
 int cpufreq_driver_target(struct cpufreq_policy *policy,
                          unsigned int target_freq,
                          unsigned int relation)
 {
-       unsigned int ret;
+       int ret;
 
        policy = cpufreq_cpu_get(policy->cpu);
        if (!policy)
                return -EINVAL;
 
-       down(&policy->lock);
+       mutex_lock(&policy->lock);
 
        ret = __cpufreq_driver_target(policy, target_freq, relation);
 
-       up(&policy->lock);
+       mutex_unlock(&policy->lock);
 
        cpufreq_cpu_put(policy);
-
        return ret;
 }
 EXPORT_SYMBOL_GPL(cpufreq_driver_target);
@@ -1151,7 +1236,7 @@ EXPORT_SYMBOL_GPL(cpufreq_driver_target);
 
 static int __cpufreq_governor(struct cpufreq_policy *policy, unsigned int event)
 {
-       int ret = -EINVAL;
+       int ret;
 
        if (!try_module_get(policy->governor->owner))
                return -EINVAL;
@@ -1177,12 +1262,11 @@ int cpufreq_governor(unsigned int cpu, unsigned int event)
        if (!policy)
                return -EINVAL;
 
-       down(&policy->lock);
+       mutex_lock(&policy->lock);
        ret = __cpufreq_governor(policy, event);
-       up(&policy->lock);
+       mutex_unlock(&policy->lock);
 
        cpufreq_cpu_put(policy);
-
        return ret;
 }
 EXPORT_SYMBOL_GPL(cpufreq_governor);
@@ -1195,18 +1279,17 @@ int cpufreq_register_governor(struct cpufreq_governor *governor)
        if (!governor)
                return -EINVAL;
 
-       down(&cpufreq_governor_sem);
-       
+       mutex_lock(&cpufreq_governor_mutex);
+
        list_for_each_entry(t, &cpufreq_governor_list, governor_list) {
                if (!strnicmp(governor->name,t->name,CPUFREQ_NAME_LEN)) {
-                       up(&cpufreq_governor_sem);
+                       mutex_unlock(&cpufreq_governor_mutex);
                        return -EBUSY;
                }
        }
        list_add(&governor->governor_list, &cpufreq_governor_list);
 
-       up(&cpufreq_governor_sem);
-
+       mutex_unlock(&cpufreq_governor_mutex);
        return 0;
 }
 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
@@ -1217,9 +1300,9 @@ void cpufreq_unregister_governor(struct cpufreq_governor *governor)
        if (!governor)
                return;
 
-       down(&cpufreq_governor_sem);
+       mutex_lock(&cpufreq_governor_mutex);
        list_del(&governor->governor_list);
-       up(&cpufreq_governor_sem);
+       mutex_unlock(&cpufreq_governor_mutex);
        return;
 }
 EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
@@ -1246,12 +1329,11 @@ int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
        if (!cpu_policy)
                return -EINVAL;
 
-       down(&cpu_policy->lock);
+       mutex_lock(&cpu_policy->lock);
        memcpy(policy, cpu_policy, sizeof(struct cpufreq_policy));
-       up(&cpu_policy->lock);
+       mutex_unlock(&cpu_policy->lock);
 
        cpufreq_cpu_put(cpu_policy);
-
        return 0;
 }
 EXPORT_SYMBOL(cpufreq_get_policy);
@@ -1265,41 +1347,33 @@ static int __cpufreq_set_policy(struct cpufreq_policy *data, struct cpufreq_poli
        dprintk("setting new policy for CPU %u: %u - %u kHz\n", policy->cpu,
                policy->min, policy->max);
 
-       memcpy(&policy->cpuinfo, 
-              &data->cpuinfo, 
-              sizeof(struct cpufreq_cpuinfo));
+       memcpy(&policy->cpuinfo, &data->cpuinfo, sizeof(struct cpufreq_cpuinfo));
 
        /* verify the cpu speed can be set within this limit */
        ret = cpufreq_driver->verify(policy);
        if (ret)
                goto error_out;
 
-       down_read(&cpufreq_notifier_rwsem);
-
        /* adjust if necessary - all reasons */
-       notifier_call_chain(&cpufreq_policy_notifier_list, CPUFREQ_ADJUST,
-                           policy);
+       blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
+                       CPUFREQ_ADJUST, policy);
 
        /* adjust if necessary - hardware incompatibility*/
-       notifier_call_chain(&cpufreq_policy_notifier_list, CPUFREQ_INCOMPATIBLE,
-                           policy);
+       blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
+                       CPUFREQ_INCOMPATIBLE, policy);
 
        /* verify the cpu speed can be set within this limit,
           which might be different to the first one */
        ret = cpufreq_driver->verify(policy);
-       if (ret) {
-               up_read(&cpufreq_notifier_rwsem);
+       if (ret)
                goto error_out;
-       }
 
        /* notification of the new policy */
-       notifier_call_chain(&cpufreq_policy_notifier_list, CPUFREQ_NOTIFY,
-                           policy);
-
-       up_read(&cpufreq_notifier_rwsem);
+       blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
+                       CPUFREQ_NOTIFY, policy);
 
-       data->min    = policy->min;
-       data->max    = policy->max;
+       data->min = policy->min;
+       data->max = policy->max;
 
        dprintk("new min and max freqs are %u - %u kHz\n", data->min, data->max);
 
@@ -1336,7 +1410,7 @@ static int __cpufreq_set_policy(struct cpufreq_policy *data, struct cpufreq_poli
                __cpufreq_governor(data, CPUFREQ_GOV_LIMITS);
        }
 
- error_out:
+error_out:
        cpufreq_debug_enable_ratelimit();
        return ret;
 }
@@ -1360,7 +1434,7 @@ int cpufreq_set_policy(struct cpufreq_policy *policy)
                return -EINVAL;
 
        /* lock this CPU */
-       down(&data->lock);
+       mutex_lock(&data->lock);
 
        ret = __cpufreq_set_policy(data, policy);
        data->user_policy.min = data->min;
@@ -1368,7 +1442,7 @@ int cpufreq_set_policy(struct cpufreq_policy *policy)
        data->user_policy.policy = data->policy;
        data->user_policy.governor = data->governor;
 
-       up(&data->lock);
+       mutex_unlock(&data->lock);
        cpufreq_cpu_put(data);
 
        return ret;
@@ -1392,26 +1466,76 @@ int cpufreq_update_policy(unsigned int cpu)
        if (!data)
                return -ENODEV;
 
-       down(&data->lock);
+       mutex_lock(&data->lock);
 
        dprintk("updating policy for CPU %u\n", cpu);
-       memcpy(&policy, 
-              data,
-              sizeof(struct cpufreq_policy));
+       memcpy(&policy, data, sizeof(struct cpufreq_policy));
        policy.min = data->user_policy.min;
        policy.max = data->user_policy.max;
        policy.policy = data->user_policy.policy;
        policy.governor = data->user_policy.governor;
 
+       /* BIOS might change freq behind our back
+         -> ask driver for current freq and notify governors about a change */
+       if (cpufreq_driver->get) {
+               policy.cur = cpufreq_driver->get(cpu);
+               if (!data->cur) {
+                       dprintk("Driver did not initialize current freq");
+                       data->cur = policy.cur;
+               } else {
+                       if (data->cur != policy.cur)
+                               cpufreq_out_of_sync(cpu, data->cur, policy.cur);
+               }
+       }
+
        ret = __cpufreq_set_policy(data, &policy);
 
-       up(&data->lock);
+       mutex_unlock(&data->lock);
 
        cpufreq_cpu_put(data);
        return ret;
 }
 EXPORT_SYMBOL(cpufreq_update_policy);
 
+static int cpufreq_cpu_callback(struct notifier_block *nfb,
+                                       unsigned long action, void *hcpu)
+{
+       unsigned int cpu = (unsigned long)hcpu;
+       struct cpufreq_policy *policy;
+       struct sys_device *sys_dev;
+
+       sys_dev = get_cpu_sysdev(cpu);
+
+       if (sys_dev) {
+               switch (action) {
+               case CPU_ONLINE:
+                       cpufreq_add_dev(sys_dev);
+                       break;
+               case CPU_DOWN_PREPARE:
+                       /*
+                        * We attempt to put this cpu in lowest frequency
+                        * possible before going down. This will permit
+                        * hardware-managed P-State to switch other related
+                        * threads to min or higher speeds if possible.
+                        */
+                       policy = cpufreq_cpu_data[cpu];
+                       if (policy) {
+                               cpufreq_driver_target(policy, policy->min,
+                                               CPUFREQ_RELATION_H);
+                       }
+                       break;
+               case CPU_DEAD:
+                       cpufreq_remove_dev(sys_dev);
+                       break;
+               }
+       }
+       return NOTIFY_OK;
+}
+
+static struct notifier_block cpufreq_cpu_notifier =
+{
+    .notifier_call = cpufreq_cpu_callback,
+};
 
 /*********************************************************************
  *               REGISTER / UNREGISTER CPUFREQ DRIVER                *
@@ -1422,9 +1546,9 @@ EXPORT_SYMBOL(cpufreq_update_policy);
  * @driver_data: A struct cpufreq_driver containing the values#
  * submitted by the CPU Frequency driver.
  *
- *   Registers a CPU Frequency driver to this core code. This code 
+ *   Registers a CPU Frequency driver to this core code. This code
  * returns zero on success, -EBUSY when another driver got here first
- * (and isn't unregistered in the meantime). 
+ * (and isn't unregistered in the meantime).
  *
  */
 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
@@ -1472,6 +1596,7 @@ int cpufreq_register_driver(struct cpufreq_driver *driver_data)
        }
 
        if (!ret) {
+               register_cpu_notifier(&cpufreq_cpu_notifier);
                dprintk("driver %s up and running\n", driver_data->name);
                cpufreq_debug_enable_ratelimit();
        }
@@ -1484,7 +1609,7 @@ EXPORT_SYMBOL_GPL(cpufreq_register_driver);
 /**
  * cpufreq_unregister_driver - unregister the current CPUFreq driver
  *
- *    Unregister the current CPUFreq driver. Only call this if you have 
+ *    Unregister the current CPUFreq driver. Only call this if you have
  * the right to do so, i.e. if you have succeeded in initialising before!
  * Returns zero if successful, and -EINVAL if the cpufreq_driver is
  * currently not initialised.
@@ -1503,6 +1628,7 @@ int cpufreq_unregister_driver(struct cpufreq_driver *driver)
        dprintk("unregistering driver %s\n", driver->name);
 
        sysdev_driver_unregister(&cpu_sysdev_class, &cpufreq_sysdev_driver);
+       unregister_cpu_notifier(&cpufreq_cpu_notifier);
 
        spin_lock_irqsave(&cpufreq_driver_lock, flags);
        cpufreq_driver = NULL;